Really outstanding article, OP. Well done. Slacked to the rest of our engineering team. We're using Google kubernetes to solve some of the fail-over/self-healing challenges you noted, and some of the other ideas as well (proxies, data store fit-to-function, etc.) but there are a couple of things in here that we either hadn't thought of (Nagle algorithm) or needed to be reminded of (timestamps at every layer, the cost of logging/eventing). Thanks for the good lunchtime read.

Thank you for this outstanding article.

Perhaps I missed it, but how exactly do you implement the timestamp logging? Do you use your normal application logging, or some other mechanism?

Some of the formatting in the blog post seems to have been lost. Words appear to start mid sentence.

Apart from that, really enjoyed the article. Would love to see more.